Impact of Host Genetics on Caecal Microbiota Composition and on Salmonella Carriage in Chicken

Background Salmonella Enteritidis (SE) is one of the major causes of human foodborne intoxication through the consumption of contaminated poultry products. Genetic selection of animals more resistant to Salmonella carriage and the modulation of gut microbiota are two promising ways of decreasing individual Salmonella carriage. This study aims to identify the main genetic and microbial factors controlling the individual levels of Salmonella carriage in chickens (Gallus gallus) in controlled experimental conditions. Two-hundred and forty animals from the White Leghorn inbred lines, N and 6 1 , were infected by SE at 7 days of age. After infection, animals were kept in isolators to reduce the recontamination of birds by Salmonella. Caecal contents were sampled at 12 days post-infection and used for DNA extraction. Microbiota DNA was used to measure individual counts of SE by digital PCR and to determine the bacterial taxonomic composition through a 16S rRNA gene high-throughput sequencing approach. Results Results conrmed that the N line is more resistant to Salmonella carriage than the 6 1 line, and that intra-line variability is higher for the 6 1 line. Furthermore, the 16S analysis showed strong signicant differences in microbiota taxonomic composition between the two lines. Out of 617 Operational Taxonomic Units (OTUs), over 390 were differentially abundant between the two lines. Furthermore, within the 6 1 line, we found a difference in the microbiota taxonomic composition between high and low Salmonella carriers, with 39 differentially abundant OTUs. Finally, via metagenome functional prediction based on 16S data, we identied several metabolic pathways potentially associated to microbiota taxonomic differences (e.g. butyrate metabolism) between high and low carriers. Conclusions Overall, this study demonstrates that the caecal microbiota composition of the N and 6 1 lines is inuenced by the host genetics, which could be one of the reasons why these lines differ for their Salmonella carriage in experimental infection conditions.


Abstract
Background Salmonella Enteritidis (SE) is one of the major causes of human foodborne intoxication through the consumption of contaminated poultry products. Genetic selection of animals more resistant to Salmonella carriage and the modulation of gut microbiota are two promising ways of decreasing individual Salmonella carriage. This study aims to identify the main genetic and microbial factors controlling the individual levels of Salmonella carriage in chickens (Gallus gallus) in controlled experimental conditions. Two-hundred and forty animals from the White Leghorn inbred lines, N and 6 1 , were infected by SE at 7 days of age. After infection, animals were kept in isolators to reduce the recontamination of birds by Salmonella. Caecal contents were sampled at 12 days post-infection and used for DNA extraction. Microbiota DNA was used to measure individual counts of SE by digital PCR and to determine the bacterial taxonomic composition through a 16S rRNA gene high-throughput sequencing approach.
Results Results con rmed that the N line is more resistant to Salmonella carriage than the 6 1 line, and that intra-line variability is higher for the 6 1 line. Furthermore, the 16S analysis showed strong signi cant differences in microbiota taxonomic composition between the two lines. Out of 617 Operational Taxonomic Units (OTUs), over 390 were differentially abundant between the two lines. Furthermore, within the 6 1 line, we found a difference in the microbiota taxonomic composition between high and low Salmonella carriers, with 39 differentially abundant OTUs. Finally, via metagenome functional prediction based on 16S data, we identi ed several metabolic pathways potentially associated to microbiota taxonomic differences (e.g. butyrate metabolism) between high and low carriers.
Conclusions Overall, this study demonstrates that the caecal microbiota composition of the N and 6 1 lines is in uenced by the host genetics, which could be one of the reasons why these lines differ for their Salmonella carriage in experimental infection conditions.

Background
Reducing Salmonella carriage in chicken ocks is a challenge to ensure human food safety and enhance breeding viability. The consumption of contaminated poultry products can indeed cause human food intoxications. The reported number of human salmonellosis outbreak cases due to the consumption of contaminated chicken meat and eggs in Europe was over 4,000 from 2014 to 2016, with more than 93% due to the Enteritidis serovar (EFSA Journal, 2019). Control of hygiene, detection of several serovars of Salmonella and culling of contaminated ocks or vaccination in farms aim to decrease this number. Nevertheless, the economic impact of human salmonellosis remains estimated at approximately 3 billion euros a year (EFSA Journal, 2019). Developing new strategies to improve the control of Salmonella propagation in poultry livestock is therefore essential to respond to this challenge.
In adult chickens, Salmonella enterica Enteritidis (SE) does not induce symptoms and can stay in the organism for a long time [1]. Carrier animals excrete the bacteria in the environment, thus increasing the bacteria propagation and facilitating the contamination of all animals. Previous studies identi ed several quantitative trait loci (QTL) associated with Salmonella carriage in the White Leghorn inbred lines N and 6 1 [2][3][4], showing the contribution of host genetic variations in the resistance to Salmonella carriage.
However, the effects associated with the identi ed loci are weak, which would imply a large polygenic control of this trait. It does not allow a direct application in commercial lines by marker-assisted selection of more resistant animals. The weak effects of the identi ed loci might also be due to a biased estimation of individual Salmonella carriage, due to the recurrent recontamination of birds by Salmonella excreted by other birds within the timeframe of experimental infections. This recontamination probably leads to a homogenization of Salmonella carriage in chicken ocks. The use of isolators, strongly reducing this recontamination, actually led to a much increased variation among birds in the experimental White Leghorn line PA12 [1], allowing the description of three categories of birds according to their shedding: super-, intermediate-and low-shedders. These isolators allow a control of the air purity and limit faecal-oral recontaminations between birds by the use of grids on the oor and decontamination of faecal drops.
Additionally, many studies demonstrate the importance of the gut microbiota for host health, in poultry as well as in every livestock species and in human. In human, disruptions of the intestinal microbiota can lead to many kinds of non-infectious diseases by altering the host physiology and metabolism and triggering in ammation [5]. It is now very clear that the host and its intestinal microbiota both contribute to the expression of many phenotypic traits of interest in livestock species [6]. In chicken, it is also well established that intestinal health is the result of complex functional interactions between intestinal microbes and host immunity [7]. In particular, the adult gut microbiota in chicken has been consistently identi ed as a protective factor to prevent the colonisation of intestines of young chicks by Salmonella sp. through a mechanism called competitive exclusion [8][9][10]. Indeed, the spraying of an adult microbiota on young chicks or the use of probiotics are already e ciently used in commercial ocks to reduce the Salmonella Enteritidis load [11][12][13][14]. Many studies have also assessed the e ciency of nutritional strategies aiming at reducing the Salmonella Enteritidis load by a modulation of the intestinal microbiota leading to an improvement of the host immunity [15][16][17][18]. Such strategies include prebiotics or probiotics boosting the production of bene cial metabolites, the modulation of host immunity, and the improvement of intestinal barrier function [19]. To understand the underlying biological mechanisms of such strategies, and to identify the bacterial taxa actually competing with Salmonella, it is important to analyse the impact of Salmonella infection on the host microbiota composition. Comparing infected and non-infected animals, such studies sought to identify OTU signatures of Salmonella infection. For example in pigs, Argüello et al. identi ed bacteria from the class Clostridia that could prevent Salmonella Typhimurium colonisation [20]. In chicken, the family Ruminococcaceae, which are more abundant in non-infected animals at 4 days post infection (dpi), could be a signature of Salmonella infection [21].
Overall, it seems both host genetic variations and intestinal microbiota composition could explain variations in individual Salmonella Enteritidis carriage in chicken. Furthermore, genetic studies on other phenotypic traits, such as digestive e ciency [22], body weight [23] or feather-pecking in laying hens [24] showed that host genetic variations in uence the intestinal microbiota composition in chicken. In these studies, authors identi ed several QTLs and single nucleotide polymorphisms (SNPs) associated with speci c bacterial species that may explain the phenotype and calculated heritabilities of these bacteria abundances.
In this study, we investigated the combined in uence of host genetic variations and gut microbiota composition on caecal Salmonella Enteritidis (SE) load. Using the chicken N and 6 1 inbred lines, we studied the impact of genetics on both the Salmonella carriage and microbiota composition at the individual level, using the isolator model, which highly decreases the exchanges of gut microbiota between animals [1]. We infected 240 animals with Salmonella Enteritidis at 7 days post hatching. We collected caecal contents at 12 dpi to assess both individual microbiota taxonomic composition and SE counts. Our aims were: (i) To assess the intra-line individual variability of Salmonella load and of the caecal microbiota composition; (ii) To compare Salmonella load and caecal microbiota composition of the two lines, in order to assess the existence of a genetic control of these two parameters; (iii) To identify putative microbial signatures of low/high Salmonella

Experimental design
Two experiments were conducted with a total of 240 animals of the White Leghorn inbred lines N and 6 1 (both males and females, with an equal distribution). For each experiment, chicks from both lines (n=120; 60 animals from each line) hatched together with free access to food and water at the experimental unit PEAT (Pole d'Expérimentation Avicole de Tours, Nouzilly, France). They were immediately transferred to the PFIE unit (Plateforme d'Infectiologie Expérimentale, INRA, Nouzilly, France) where they were reared together in battery cages. On day 7, each chick was orally infected with Salmonellaenterica Enteritidis (Strain 775 [LA5 Nal20Sm500], 5.10 4 cfu/0.2 mL/chick) and animals were transferred into four isolators to decrease oro-faecal recontaminations as described previously [1]. Two isolators contained chicks from line N and two other isolators contained chicks from line 6 1 , with 30 birds per isolator. Sibs and half-sibs were separated into the two isolators available for each line, in order to prevent confounding between an intra-line genetic effect on Salmonella carriage and an effect of the isolator. Isolators with birds from line 6 1 in the rst experiment contained birds from line N in the second experiment, in order to prevent confounding between isolator and line effects. Then, 12 days post-infection, animals were euthanized with CO2 according to the French regulation for experimental chickens and caecal contents were gently collected so as not to remove the intestinal mucosa, weighted, transferred in cryotubes and immediately frozen in liquid nitrogen and stored at -80°C until use. All animal procedures has been approved by the Ethic committee (APAFIS#5833-2016062416362298v3) and authorised by the French Government.

DNA extraction
Individual caecal DNA was extracted from an average of 200 mg of frozen caecal contents as previously described [25]. In brief, samples were incubated at 70°C for 1 hour with 250 μl of guanidine thiocyanate buffer (4 M guanidine thiocyanate-0.1 M Tris (pH 7.5) and 40 μl of 10% N-lauroyl sarcosine-0.1 M phosphate buffer (pH 8.0)) and 500 μl of 5% N-lauroyl sarcosine. One volume (750 μl) of 0.1-mmdiameter silica beads (Sigma) was added, and tubes were shaken for 10 min at the maximum speed of a MM200 Mixer Mill (Retsch, Germany). Tubes were vortexed and centrifuged at 14000 rpm for 5 min at 4°C. After recovery of the supernatant, 30 μl of Proteinase K (Chemagic STAR DNA BTS kit, Perkin Elmer, USA) were added and samples were incubated for 10 min at 70°C at 250 rpm in Multi-Therm (Benchmark Scienti c, USA), then for 5 min at 95°C for enzyme inactivation. Tubes were centrifuged at 14000 rpm for 5 min at 4°C and supernatants were transferred into a deepwell plate. The plate was loaded on the nucleic acid workstation Chemagic STAR (Hamilton, Perkin Elmer, USA) and the extraction protocol was performed with Chemagic STAR DNA BTS kit (Perkin Elmer, USA) by the @BRIDGe platform (INRAE, Jouyen-Josas, France) according to the manufacturer's instructions. DNA concentration was measured by uorometric quanti cation (Qubit) and DNA samples were stored at -20°C.
Quanti cation of S. Enteritidis by ddPCR Individual abundances of Salmonella Enteritidis in caecal contents at day 19 were obtained by Droplet Digital PCR using the QX200 Droplet Digital PCR system (Bio-Rad) at the @BRIDGe platform (INRAE, Jouy-en-Josas, France). Each DNA sample was diluted at 1:2 and 1:5 in a nal volume of 5 µl to be under PCR saturated conditions. Droplets were generated from the 20 µl of the total volume of ampli cation mix and genomic DNA (15000 to 20000 droplets per sample). The ampli cation targeted a region of the InvA gene speci c to Salmonella Enteritidis using speci c primers, as described by Daum et al. [26] (Additional le 1). PCR cycling condition started with an enzyme activation step at 95°C for 10 min, followed by 40 cycles at 94°C for 30 min and 60°C for 1 min, and ended by an enzyme deactivation step at 98°C for 10 min. For all cycling steps, we used a 2.5°C/sec ramp rate.
For each sample, the number of copies of Salmonella per gram of caecal content was calculated from the average number of copies of Salmonella per µl of the two dilutions, assuming that each ampli ed copy of the InvA gene corresponds to one Salmonella bacterium. Data were analysed with a log transformation of the copies of Salmonella. Analyses of variance (Anova) were performed to test the signi cance of differences of the copies of Salmonella between conditions (line, sex, experiment or isolator) using the anova function in R software version 3.5.1 (Type I sum of squares).

PCR and Sequencing of 16S rRNA Genes
Ampli cation of the V3-V4 hyper-variable region of the 16S rRNA coding gene was performed at the INRAE @BRIDGe platform. Universal V3-V4 primers (Additional le 1) were used for the rst PCR reaction. PCR cycling conditions were as follows: an initial denaturation step (94°C for 10 min), 35 cycles of ampli cation (94°C for 1 min, 68° C for 1 min and 72°c for 1 min) and a nal elongation step at 72°C for 10 min. Amplicons were then puri ed using magnetic beads (Clean NA, GC biotech B.V., The Nederlands) and the concentration was controlled using a Nanodrop spectrophotometer (Thermo Scienti c, USA). In the second PCR, samples were multiplexed and the second pair of primers was used (Additional le 1). An initial denaturation step (94°C for 10 min), 12 cycles of ampli cation (94°C for 1 min, 65°C for 1 min and 72°C for 1 min) and a nal elongation step at 72°C for 10 min were performed. Amplicons were puri ed and concentration was controlled as described for the rst PCR reaction. One run on an Illumina MiSeq was used to sequence amplicons (2 x 250 paired-end reads) according to the standard protocol.

Bioinformatic and statistical analyses
Identi cation of Operational Taxonomic Units (OTUs) was performed by using the FROGS pipeline [27]. FastQC program to control quality and the Cutadapt program to nd and remove adapter sequences from sequencing reads, and R1 and R2 reads were merged and ltered (at Phred >= Q20) by using the Flash program [28]. OTUs were identi ed using the Swarm program [29]. Chimera OTUs were removed using the VSEARCH program [30], taxa ltering was performed with a minimum abundance threshold of 0.005% as proposed by [31]. Finally, phylogenetic a liations were identi ed using the Silva database by using the blastn+ program [32]. OTUs representing less than 0.5% of global reads and samples with less than 10,000 reads were removed.
The phyloseq (1.24.2) and vegan (2.5-3) packages were used with R to perform the diversity analysis on normalized data. Alpha diversity was measured using the Shannon index and beta diversity using the Whittaker index. Analyses of variance (Anova) were performed to test the signi cance of differences of alpha and beta diversity between conditions (line, sex, experiment or isolator). We evaluated the Bray-Curtis distances using the env_ t method that we plotted in an NMDS (Non-Metric Multidimensional Scaling) representation. Permutational Multivariate Analyses of Variance (Permanova) were used to test for the signi cance of differences of diversity according to variations in lines, experiments, isolators and sex. The R package metagenomeSeq (1.24.1) was used for the identi cation of differentially abundant (DA) OTUs between lines N and 6 1 and between high and low carriers within line 6 1 . The OTU table was rst normalised, and the model was tted using the tZIG method by including experiment, isolator and sex as cofactors. Heatmaps were built using the function plotMRheatmap on signi cant DA OTUs. Then, OTUs were aggregated by family and genus with the command aggregateByTaxonomy to identify DA families and genera. Functional gene families and MetaCyc pathway were predicted using the PICRUST2 package. DA KEGG ortholog (KO) and DA MetaCyc pathway were identi ed with the R package DESeq2 (1.26.0). DA MetaCyc pathways were aggregated at the super-pathway level using MetaCyc database [33].
Metadata table, unrare ed OTU table, and corresponding taxonomic classi cations have been included as Additional les 2, 3 and 4, respectively.

Abundance of S. Enteritidis in caecal contents
We obtained an average of 3.38 log10 copies of DNA target sequences per gram of caecal content for line 6 1 , and 2.55 log10 copies for line N ( Figure 1). An Anova including genetic line, experiment, isolator and sex as factors showed a signi cant difference (P<0.001) of ISC (Individual Salmonella Carriage) between the two lines ( Table 1 and Additional le 5, Table 1). This difference between lines was also signi cant when considering each experiment separately (Table 1). We also observed that ISC is more variable in line 6 1 than in line N (P<0.001), with standard deviations of 1.3 and 0.7, respectively. This higher ISC variation in line 6 1 was more marked in experiment 1 than in experiment 2 (P<0.001), with standard deviations of 1.61 vs. 0.84 ( Figure 1). ISC was not signi cantly different between males and females (Additional le 5, Table 1). Furthermore, the mean ISC between the two experiments was signi cantly different (P<0.001) for line N (2.2±0.7 in experiment 1 vs. 2.9±0.4 in experiment 2) but not for line 6 1 . Finally, within each experiment, there were no signi cant differences between isolators containing line 6 1 , but there were signi cant differences between isolators containing line N (P<0.01). percentage of a liation of 99% at the family level and 65% at the genus level. Samples with a total number of reads lower than 10,000 were removed, so that 182 samples from the 228 sequenced samples were nally analysed: 86 samples from line 6 1 and 95 from line N.
At the phylum level, Firmicutes, followed by Proteobacteria and Bacteroidetes (Figure 2), dominated the bacterial composition. At the family level, we observed a predominance of Lachnospiraceae and Ruminoncoccaceae, followed by the Enterobacteriaceae, Clostridiales and Bacillaceae (Figure 2 and Additional le 5, Table 2).
The analysis of alpha diversity with the Shannon index did not show a signi cant difference between lines, experiments, or sex (Additional le 5, Table 3). Using the Whittaker index, we observed a signi cant difference of beta diversity between experiments (P<0.0001), corresponding to 0.24 for experiment 1 vs. 0.20 for experiment 2. However, we did not observe any difference in beta diversity between lines (0.20 for both). Comparing high-and low-carriers in line 6 1 , we identi ed no signi cant difference in alpha diversity and richness, but we identi ed a signi cant difference in beta diversity (Additional le 5, Table 3).

Non-metric multidimensional scaling and differentially abundant OTUs between lines
The analysis of the beta diversity using the Bray-Curtis distance and an NMDS representation ( Figure 3) showed that the microbiota composition is clearly clustered according to the genetic line. This clustering is observed for each experiment individually but is more contrasted in experiment 1 ( Figure 3). The Permanova analysis con rmed this line effect (P<0.001) after correcting for the experiment, isolator and sex effects, whether merging data from the two experiments or considering each experiment separately (Additional le 5, Table 4). The global microbiota composition was also signi cantly different between experiments and isolators, but not signi cantly different between males and females (Additional le 5, Table 4).
Of the 617 OTUs identi ed, we observed a total of 390 signi cant DA OTUs between lines over the two experiments, after controlling for the isolator effect (Additional le 5, Table 5). When analysing experiments separately, a heatmap showed a more strongly de ned clustering for experiment 1, with 388 DA OTUs between lines (Figure 3), whereas for experiment 2, we observed 284 DA OTUs. A total of 187 DA OTUs were found to be in common between the two experiments.
In the following analyses, we aggregated the 617 OTUs in order to compare compositions at higher taxonomic ranks, which led to 14 families and 51 genera. We identi ed 9 DA families and 31 DA genera between lines over the two experiments, after adjusting for the isolator effect ( Figure 4 and Additional le 5, Table 6). For experiment 1 and experiment 2, we respectively identi ed 9 and 5 DA families and 24 and 29 DA genera (Figure 4) High and low carriers in line 6 1

and the family Christensenellaceae
In order to assess a potential difference of microbiota composition between high and low carriers in line 6 1 , we used an NMDS representation for the rst experiment (Figure 4, A). We chose experiment 1 because it had the largest variability in individual Salmonella carriage, which allowed for the identi cation of high and low carriers. The Permanova analysis showed a signi cant (P<0.01) difference of microbiota composition between both groups. The metagenomeSeq analysis led to the identi cation of 39 DA OTUs between high and low carriers (Figure 4, B and Additional le 5, Table 7). After aggregation, we obtained one DA family: Christensenellaceae, and three DA genera: Ruminococcaceae NK4A214 group, Ruminiclostridium 5 and Christensenellaceae R-7 group ( Figure 5, C and Additional le 5, Table 8).
A comparison of the results in DA analyses between lines and between high and low carriers led to the identi cation of a common, DA family: the Christensenellaceae. Christensenellaceae is more highly abundant in low carriers than in high carriers in line 6 1 and is also more highly abundant in line N, which is more resistant (i.e. low carrier) to Salmonella. The same observation can be made for the DA genus Christensenellaceae R7-group, which is more abundant in low carriers as well as in the resistant line N for both experiments. A regression analysis led to a signi cant correlation (P-value = 4.11e -06 ) between ISC and Christensenellaceae abundance. high carriers. Finally, through our caecal microbiota taxonomic analysis, we identi ed taxa and metabolism pathways that may be associated with Salmonella carriage.
In both experiments conducted, line N was more resistant and microbiota taxonomic composition was clearly different between both lines. Nevertheless, ISC was higher for line N in the second experiment and the DA OTUs between lines were not strictly the same between experiments. These differences in microbiota composition might be due to the impact of the immediate environment on individual microbiota composition. Although conditions were controlled to be similar between both experiments, even slight changes might lead to differences in the primo-colonisation of the intestinal tract [34,35]. The hatching environment might have differed between both batches and/or the two batches of eggs might have carried different microbes. Those potential differences might have affected the bacterial gut primocolonisation of the newborn chicks. As a result, a difference in microbiota composition before infection could affect the microbiota composition after infection. In the following section, since ISC and microbiota composition differed slightly between experiments, we will focus only on results validated in both experiments.

Impact of the chicken genetic line on the Individual Salmonella Enteritidis carriage
Previous studies showed the impact of host genetics on Salmonella carriage, using different infection models. Signi cant differences in Salmonella Enteritidis carriage in commercial or local chicken breed have already been identi ed [36][37][38]. Our results with lines N and 6 1 con rm previous results obtained without using isolators with the same lines, with on-oor grouped rearing [2,39], con rming that line N is a lower Salmonella Enteritidis carrier than line 6 1 . Previous studies validated candidate genes associated with resistance to Salmonella Enteritidis carriage, such as SLC11A1 and TLR4 [40][41][42] and identi ed QTLs on several chromosomes i [2][3][4]. Nevertheless, due to the highly polygenic control of carrier-state in these lines, with many loci with weak effects involved [43], identi cation of causal genes at the QTLs has not been yet possible. The infection of birds in isolators, which would allow for much larger individual variability [1], might be a way to improve future QTL detection studies and identi cation of causal genes. Interestingly, we observed a larger variability in ISC in line 6 1 , con rmed in two independent experiments.
This larger variability of ISC generates new hypotheses: could this ISC variability be caused by intra-line genetic variations, or by a variability of the microbiota composition?

Impact of the genetic line on the microbiota composition
In both experiments conducted, we showed a clear difference of the caecal microbiota composition between lines N and 6 1 at 12 days post-infection, which suggests the existence of a host genetic control of this composition. For each experiment, birds from lines N and 6 1 were hatched in the same environment at the same time, and raised together until the experimental infection, so that the initial microbial environment was similar for all birds tested before infection. Differences in caecal microbiota composition between lines can therefore not be attributed to differential exposures to environmental microbes before infection. Furthermore, for experiment 1, we observed no signi cant difference of microbiota composition between the two isolators used for the same line (P>0.1, Additional le 5, Table  4). Thus, we cannot associate differences in microbiota composition between lines with isolators in experiment 1. In experiment 2, we observed a signi cant difference of microbiota composition between the two isolators used for the same line (P<0.01, Additional le 5, Table 4). It can be argued that isolators might be populated by different microbial populations, which could in turn in uence the caecal microbiota composition after infection. Howerver, isolators were sterilized between each experimentation. In addition, we inverted the isolators containing each line in the second experiment, and still observed a similar difference between lines in terms of caecal microbiota composition. Moreover, since the utilisation of isolators decreases the oro-faecal recontamination of commensal microbiota as well as pathogenic bacteria, the caecal microbiota composition of each chicken can mature isolated from the impact of the other birds. In chicken as well as in other livestock species and in human, the genetic control of the intestinal microbiota composition is well documented. In human studies, the high sensitivity of the gut microbiota to a myriad of parameters, especially differences in diet, makes statistical analyses complex due to many confounding factors. In spite of these di culties, some bacteria known to be associated with immunity seem to be heritable and candidate genes have been identi ed [47]. For example, the family Christensenellaceae, which are the most heritable bacteria in the study of Goodrich et al. in human, show a heritability of 0.39, while other studies reported an heritability of 0.62 for this family [48,49]. Candidate genes in human that might be associated with Christensenellaceae abundance, such as ILR23 or FUT2,were also identi ed [50,51]. In chicken, studies identi ed differences in microbiota composition between genetic lines, for instance in two chicken lines that differ in their susceptibility to bacterial infections [52] or in two divergent genetic lines for body weight [53]. Another study compared four commercial lines and an indigenous Indian breed, which revealed a signi cant impact of the genetic background on microbiota composition and led to the identi cation of 42 speci c biomarkers [54]. At least two studies identi ed moderate heritabilities of bacteria families and several QTLs involved in the control of these bacterial abundances [22,55].
Finally, we conclude that this clear difference of microbiota composition between lines after Salmonella infection is probably caused by host genetic variations between lines N and 6 1 . We also showed a difference in ISC between lines. Therefore, in contrast to Chintoan-Uta's conclusion for Campylobacter, we formulate the hypothesis that genetic variations between lines N and 6 1 might indirectly in uence Salmonella carriage through an in uence on the microbiota composition. This does not exclude other potential pathways, in particular those involving host immunity. For example, the candidate genes SIVA1, implicated in a cell death mechanism in extracellular trap production, or SLC11A1, associated with heterophil extracellular trap production and phagocytosis of SE, may be implicated in leukocyte function and in the inhibition of intracellular bacterial growth by depleting metal ions [56].

Relation of caecal microbiota diversity and composition with individual Salmonella Enteritidis carriage
The colonisation and the adhesion of commensal bacteria covering the mucosal epithelium constitute a protective bio lm by their competitive exclusion (CE) function. Studies a rm that CE is currently the best approach to decrease Salmonella colonization in chicken in commercial production [9].
In our study, all animals were experimentally infected and the identi cation of signature bacteria was performed by comparing chickens with differences of individual Salmonella carriage between lines or intra-lines. Our hypothesis is that the abundance of commensal, potentially competitive bacteria is higher in the microbiota of resistant chickens, thus preventing the colonisation of Salmonella in the intestinal tract. The genetic background could in part control this abundance of competitive bacteria. We rst compared the microbiota of line 6 1 (susceptible) to line N (resistant), and we subsequently compared the high and low carriers in line 6 1 . Thus, differences of caecal microbiota diversity or composition between lines and between high and low carriers could be indicative of potential signature bacteria taxa of high and low ISC.
The weak correlation of microbiota diversity with ISC The average value of microbiota beta diversity within a group measures the similarity of microbiota composition of each chicken of this group compared to all the chickens of the group. The higher the value of beta diversity, the more individual microbiota in the group differ from each other. We compared here two groups: line N vs. line 6 1 .
We showed that the microbiota beta diversity is not signi cantly different between lines (Whittaker index = 0.2 for the two lines). Thus, the difference of ISC between lines and the larger ISC variability in line 6 1 cannot be related to a difference in beta diversity of the caecal microbiota. Similarly, the average alpha diversity was not signi cantly different between lines and cannot be related to the observed differences of ISC. Thus, at 12 dpi, it seems that differences in genetic backgrounds do not have an impact on the microbiota diversity. Furthermore, the comparison of high and low carriers in line 6 1 showed that alpha diversity cannot be associated with differences in ISC. Comparing chickens infected and non-infected with SE, Liu and colleagues (2018) showed a slight decrease of richness at 14 dpi in the infected group, but no signi cant difference for the Shannon index [44]. At 10 dpi, another study similarly showed no signi cant difference between infected and non-infected groups [21]. These observations corroborate ours and lead us to conclude that the infection with SE or the level of SE carriage do not affect the OTU diversity in the caecal microbiota.
However, we identi ed a signi cant difference in beta diversity between high and low carriers within line 6 1 : individual microbiota of high carriers are more similar compared to the individual microbiota of low carriers, which harbour more differences. Does the higher level of Salmonella drive the microbiota to a more similar composition in high-carriers? Or are these animals more susceptible to Salmonella because some shared characteristics of their microbiota lead to a less e cient competitive exclusion? These questions remain open; as we also cannot exclude that genetic variations within line 6 1 could explain these differences in caecal microbiota beta diversity.
Correlation between individual microbiota composition and Individual Salmonella Enteritidis carriage and functional analysis DA OTUs and pathways identi ed between lines could potentially be associated with differences in ISC. Likewise, DA OTUs and pathways identi ed between high and low carriers in line 6 1 could be associated with ISC.

Short-Chain Fatty Acids (SCFAs) metabolic pathway
We showed that the SCFAs metabolic pathway would be more abundant both in the resistant line N and in low carriers in the susceptible line 6 1 . Thus, the production of SCFAs could be associated with low Salmonella carriage. This result is coherent with many studies showing a relationship between Salmonella colonisation and SCFA. Many studies showed that the production of SCFAs, such as butyrate, by the intestinal microbiota has bene cial effects for the host. Studies have conferred to SCFAs several functions: the regulation of immune cell function [57], the activation of macrophages [58] or the maintenance of the oxygen balance preventing dysbiosis [59]. In mice, Zhou and collaborators showed in 2017 that butyric acid could increase the host capacity to resist to pathogen infection by restoring the gastrointestinal barrier [60], and in 2019, they showed that butyric acid can decrease the in ammation in chicken [61]. Besides, we know that an increase of in ammation makes electron acceptors available to Salmonella for oxygen respiration, which in turn increases the ability of Salmonella for competition and colonisation [62]. More speci cally, it was shown that the rise of SCFAs on in vitro culture of avian intestinal cells decreases the pathogenicity of Salmonella, blocking its entry into the organism [63]. Thus, line N and low carriers in line 6 1 could carry a bene cial microbiota for Salmonella resistance.

Catechol degradation pathway
The catechol degradation pathway was also more abundant in both the resistant line N and in the low Salmonella carriers of the susceptible line 6 1 . Several studies report the association of catechol with Salmonella virulence. Salmonella have the ability to produce auto-inducers 3 (AI-3), which have a similar chemical structure as the catecholamine from the catechol family [64,65]. In chicken and in pigs, studies have already showed that a treatment with catecholamine, e.g. Norepinephrine, increases Salmonella colonization in the host and Salmonella spread in the environment [66,67]. Two mechanisms have been described to explain this phenomenon: the iron availability and the quorum sensing signal. A study in mice on Salmonella Typhimurium showed that molecules from the catechol family increase the iron availability for Salmonella by chelating iron III in their aromatic function and increasing iron accumulation in macrophages, which in turn facilitate Salmonella colonisation [68]. Besides, catechol plays a role of quorum sensing signal for the production of bio lm, thus increasing the virulence of Salmonella during host infection [64,69]. Outside the host, this bio lm increases the Salmonella capacity to resist on eggs or meat, rising salmonellosis risks in human [70]. Thus, the microbiota capacity of catechol degradation in line N and low carriers in line 6 1 could be bene cial for Salmonella resistance.

Family Christensenellaceae
We showed that the Christensenellaceae family is more abundant in low Salmonella carriers ( Figure 4). This bacteria family was already associated with bene cial impact on health in human and in mice [49]. For example, it has been associated with longevity [71], with bene cial impacts on obesity [48,72] and with metabolic health [49]. The ability of Christensenellaceae to produce butyric acid confers to these bacteria a real health interest (refer to the section "Short-Chain Fatty Acids metabolic pathway"). Interestingly, in parallel, it was shown that Christensenellaceae is one of the most heritable bacterial families of the human intestinal microbiota [48]. This leads us to formulate the hypothesis that host genetic variations between lines N and 6 1 might cause variations in Christensenellaceae abundance, which in turn could affect Salmonella resistance. However, to our knowledge, the heritability of Christensenellaceae has not been assessed in chicken. Finally, Azcarate-Peril et al. showed that the use of Galacto-Oligosaccharides (GOS) as a prebiotic in chicken increases the clearance of Salmonella Typhimurium after infection and interestingly also increases the Christensenellaceae abundance [73]. Indeed, Christensenellaceae have the capacity to metabolize GOS unlike the host or Salmonella, which could be in disfavour of Salmonella [73,74].
Family Enterobacteriaceae, Butyrate producers and anaerobisation The family Enterobacteriaceae was more abundant in the resistant line ( Figure 4). These bacteria are in competition with Salmonella for the respiration of oxygen [75] and for the use of nutrients as iron [76]. These bacteria are also able to product bacteriocine to inhibit the proliferation of Salmonella [75]. The family Ruminococcaceae and the genera Flavonifractor, Pseudo avonifractor, Anaerostipes and Intestinimonas, which are butyrate producers [77,78], were also more abundant in the resistant line. We have already described the bene cial effects of butyrate (see the section "Short-Chain Fatty Acids metabolic pathway "). We can add that, according to Litvak and colleagues in 2019, the combined effects of butyrate production, maintaining a low oxygen concentration, and oxygen respiration by competitive bacteria such as Enterobacteriaceae, lead to an anaerobisation of the lumen and thus, a decrease of the capacity of Salmonella colonisation [75]. Our results are compatible with this hypothesis.

Family Ruminococcaceae and in ammation
We showed that the Ruminococcaceae family is less abundant in the susceptible line 6 1 (Figure 4). It has been shown that a decrease of Ruminococcaceae can be associated with an increase of in ammation [78] and thus an increase of Salmonella competition [62].

Other bacteria
Other interesting bacteria associated with low carriage were identi ed in experiment 2 but not con rmed in experiment 1. This is the case of Lactobacillus, which were more abundant in the resistant line. Some species of these bacteria, used as a probiotic, were shown to signi cantly decrease the Salmonella Enteritidis carriage [79,80] and are also associated with the acceleration of Salmonella Typhimurium clearance in chicken [73]. Nevertheless, bacteria bene cial for health have also been found with a higher abundance in the susceptible line. For example, the Blautia genus, which is more abundant in line 6 1 in experiment 2 (not con rmed in experiment 1), is a butyrate producer [77] and has bene cial antiin ammatory effects [81]. Thus, looking at individual taxa might not be su cient. More likely, we suggest that the total balance of bene cial bacteria has an impact on the Salmonella resistance, which supports the idea of studying the aggregated contributions of several taxa to the same metabolic pathway.

Conclusions
We showed an impact of the genetic line on individual Salmonella Enteritidis count in caeca and on caecal microbiota taxonomic composition. We also showed associations between the abundances of bacterial taxa and metabolic pathways with the ISC status (high vs. low carriers), which were previously associated with resistance to Salmonella Enteritidis. Most notably, we identi ed an overrepresentation of the Short-Chain Fatty Acids metabolic pathway and the Catechol degradation pathway, as well as the Christensenellaceae, Enterobacteriaceae and Ruminococcaceae families in low-carriers. Based on these observations, we hypothesize that genetic differences between lines N and 6 1 may in uence the level of Salmonella carriage by in uencing the abundances of bene cial bacteria. Combining information on host genetics and gut microbiota composition is useful to sharpen the prediction of complex traits such as resistance to pathogens. Our study showed that both caecal microbiota and the host genetic background play a role in the mechanisms leading to Salmonella colonization resistance in chickens. Future studies should decipher the genes that potentially control differences in bacterial abundances in these lines. All animal procedures has been approved by the Ethic committee (APAFIS#5833-2016062416362298v3) and authorised by the French Government.

Consent for publication
Not applicable Availability of data and material Sequencing data analysed during the current study are available in the NCBI Sequence Read Archive (SRA) database under the Bioproject accession number PRJNA649900. All data generated and analysed during this study are included in this published article and its supplementary information les.

Competing interests
The authors declare that they have no competing interests. Salmonella Enteritidis abundance at 12 dpi in caecal contents. Salmonella Enteritidis abundance at 12 dpi in caecal contents of chickens from lines N and 61 infected with S. Enteritidis (log10/g of caecal contents) according to the experiment and the isolator. We observe a signi cant difference of Salmonella carriage between lines and between experiments. In line 61, the carriage is more variable between chicks, particularly for the experiment 1.